Randomness in computer systems

Most randomness needed in practical computer systems is generated purely by programs, as discussed on page 317. But to avoid having a particular program give exactly the same random sequence every time it is run, one usually starts from a seed chosen on the basis of some random feature of the environment. Until the early 1990s this seed was most often taken from the exact time of day indicated by the computer's clock at the moment when it was requested. But particularly in environments where multiple programs can start almost simultaneously other approaches became necessary. Versions of the Unix operating system, for example, began to support a virtual device (typically called /dev/random) to maintain a kind of pool of randomness based on details of the computer system. Most often this uses precise timings between interrupts generated by keys being pressed, a mouse being moved, or data being delivered from a disk, network, or other device. And to prevent the same state being reached every time a computer is rebooted, some information is permanently maintained in a file. At the end of the 1990s standard microprocessors also began to include instructions to sample thermal noise from an on-chip resistor. (Any password or encryption key made up by a human can be thought of as a source of randomness; some systems look at details of biometric data, or scribbles drawn with a mouse.)